Technical University of Crete
Department of Electronic and Computer Engineering
REQUIREMENTS ANALYSIS FOR MEDICAL INFORMATION
SYSTEMS DESIGN
by
RENA PERAKI
Chania, March 2008
ii
Abstract
The World Wide Web (the Web) is an increasingly popular source for health
information. Also, vast amounts of health information are available in medical data
bases. Making this information available in a concise and uniform way is of crucial
importance to individual users or organizations. Users of the medical domain can be
either health care professionals (experts) or consumers (novice users). Consumers
searching for medical information usually issue a natural language query and need to
retrieve medical documents on a topic of interest that is easy to read and comprehend
(e.g. medical documents that don’t contain complex medical terminology). On the
other hand, experts usually do more specialized searches (involving complex
terminology) looking for state-of-the-art documents to fulfill their information needs.
This thesis deals with issues related to the design and implementation of medical
information systems to fulfill the need of both types of users. Medical information is
acquired by web sources or by data repositories available (e.g. Medline). Acquired
medical documents from authoritative (validated) sources are analyzed (though text
analysis) and automatically indexed by author, type, content and association (e.g.
link) information. System’s functionality is described using use cases tools and the
Unified Modeling Language (UML). UML case, package, activities and class
diagrams are used to model users, data entities, system functionalities and their interrelationships. Also part of the system’s functionality has been implemented in a
system prototype.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
iii
Περίληψη
O συνεχώς αυξανόμενος όγκος της ιατρικής πληροφορίας που είναι διαθέσιμος τόσο
στο διαδίκτυο όσο και σε ιατρικές βάσεις δεδομένων, έχει αυξήσει σημαντικά το
ενδιαφέρον για συστήματα που παρέχουν αυτή την πληροφορία. Οι χρήστες τέτοιων
συστημάτων μπορεί να είναι είτε επαγγελματίες από το χώρο της παροχής
υγειονομικής φροντίδας (γιατροί, φαρμακοποιοί, νοσηλευτές, φοιτητές τμημάτων
ιατρικής ειδικότητας κλπ), είτε απλοί χρήστες που αναζητούν πληροφορία για κάποια
προσωπική υπόθεση υγείας, για πρόληψη κλπ. Οι ανάγκες αυτών των δύο ομάδων
χρηστών είναι διαφορετικές με κοινό στοιχείο τον εντοπισμό ιατρικών κειμένων από
έγκυρες και αξιόπιστες πηγές. Οι απλοί χρήστες, σε αντίθεση με τους επαγγελματίες
στο χώρο της υγείας,
επιθυμούν να εντοπίζουν ιατρικά κείμενα σε κατανοητή
γλώσσα, χωρίς πολύπλοκους ιατρικούς όρους. Σε αυτή την εργασία παρουσιάζεται
ένα σύστημα παροχής ιατρικής πληροφορίας που προσπαθεί να καλύψει τις ανάγκες
και των δύο διαφορετικών τύπων χρηστών. Τα εργαλεία που χρησιμοποιούνται για
την καταγραφή των απαιτήσεων και τον σχεδιασμό του συστήματος είναι οι
περιπτώσεις χρήσης (use cases) και τα διαγράμματα της γλώσσας μοντελοποίησης
UML. Πιο συγκεκριμένα χρησιμοποιούνται τα διαγράμματα περιπτώσεων χρήσης
(use case diagrams), πακέτου (packet diagrams), δραστηριοτήτων (activities
diagrams) για τον σχεδιασμό των διαδικασιών που θα ακολουθηθούν στην υλοποίηση
και τέλος τα διαγράμματα κλάσεων (class diagrams) για τον σχεδιασμό των
δεδομένων του συστήματος. Ένα μέρος αυτής της λειτουργικότητας έχει υλοποιηθεί
και παρουσιάζεται στην πρώτη έκδοση του συστήματος.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
iv
Αυτή η εργασία αφιερώνεται στη
μνήμη του αγαπημένου μου πατέρα Στέλιου …
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
v
Acknowledgements
I would like to sincerely thank my supervisor Professor Euripides Petrakis, for his
assistance, support and the encouragement he showed to me. I thank also the other
members of my thesis committee, Professors Stavros Christodoulakis and Michail
Lagoudakis for their helpful suggestions. In addition, I am grateful to my colleague in
the Intelligent Laboratory of Technical University of Crete Angelos Hliaoutakis for
his collaboration and for the valuable comments he provided.
I offer special thanks to my parents Stelios and Gitsa, my husband Costis Halkiadakis
and of course my children Stela, Dimitris and Stelios. I thank them for their love and
encouragement in all aspects of my life.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
vi
Table of Contents
Chapter 1 - INTRODUCTION
1.1
1.2
1.3
THE SYSTEM
CONTRIBUTIONS
THESIS STRUCTURE
2
3
4
Chapter 2 - BACKGROUND AND RELATED WORK
2.1
2.2
2.3
2.4
2.5
2.6
2.7
UML (Unified Modeling Language)
USE CASES
MeSH
UMLS
MMTx
AMTEx
RELATED SYSTEMS
6
9
11
12
13
13
14
Chapter 3 - REQUIREMENTS ANALYSIS
3.1
3.2
3.3
3.4
3.5
3.5.1
3.5.2
3.5.3
3.5.4
PACKAGE DIAGRAM
USE CASE DIAGRAM
USE CASES
ACTIVITY DIAGRAMS
CLASS DIAGRAMS
Class diagram#1 – SYSTEM MANAGEMENT PACKAGE
Class diagram #2 – DOCUMENT RETRIEVAL PACKAGE
Class diagram #3 – USER MANAGEMENT PACKAGE
Class diagram #4 – USER INTERFACE PACKAGE
Chapter 4 - IMPLEMENTATION
GLOSSARY
REFERENCES
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
17
19
21
52
70
71
73
75
76
78
81
83
vii
List of Figures
Figure 3.1
Package diagram
18
Figure 3.2
Use case diagram
20
Figure 3.3
The system’s architecture
21
Figure 3.4
Usecase’s hierarchical structure
22
Figure 3.5
Activity diagram#1 – Use the application
53
Figure 3.6
Activity diagram#2 – User registration
54
Figure 3.7
Activity diagram#3 – User login
55
Figure 3.8
Activity diagram#4 – System management
56
Figure 3.9
Activity diagram#5 – Create the System’s database
57
Figure 3.10
Activity diagram#6 – Update the System’s database
58
Figure 3.11
Activity diagram#7 – Build citation indexes
59
Figure 3.12
Activity diagram#8 – Extract the document’s terms and categorize
60
the document
Figure 3.13
Activity diagram#9 – Create the representation vectors for the
61
document or the user’s query
Figure 3.14
Activity diagram#10 – Categorize the document to
62
consumer/expert category
Figure 3.15
Activity diagram#11 – Build indexes for the document’s terms
63
Figure 3.16
Activity diagram#12 – Answer the query
64
Figure 3.17
Activity diagram#13 – Retrieve relevant documents
65
Figure 3.18
Activity diagram#14 – Rank the documents
66
Figure 3.19
Activity diagram#15 – Present results to user
67
Figure 3.20
Activity diagram#16 – Reformulate the user’s query
68
Figure 3.21
Activity diagram#17 – Browse database by medical topics
69
Figure 3.22
System Management Package – Class diagram
72
Figure 3.23
Document Retrieval Package – Class diagram
74
Figure 3.24
User Management Package – Class diagram
75
Figure 3.25
User Interface Package – Class diagram
77
Figure 4.1
User registers into the system
79
Figure 4.2
User submits a free text query
80
Figure 4.3
Results presentation
80
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
1
Chapter 1
INTRODUCTION
The amount of medical information available on the Internet is huge and it is growing
every day. Recent surveys indicate that 80 percent of adult Internet users, say they
have searched at least one health topic at some point [15]. Usually, the users are
uncertain about their exact questions and unfamiliar with the medical domain and the
medical vocabulary (called consumers). Of course, there are health care professional
users (the experts), who use medical terminology for their search and have extra
needs, such as searching the literature by following citations in articles.
To facilitate users to acquire medical information from the Web, many medical
information systems have been developed, but medical information retrieval remains
difficult for both consumers and experts. The problem that consumers usually face in
using existing medical search engines is that the available health information is most
of the times ‘not friendly’ to them, complex and full of medical terminology. Another
problem is that when the consumer enters a medical topic, many medical web sites
use similar, but not identical phrases for the same topic.
Another critical issue in designing these systems is not just to find information, but to
locate the right piece of information, especially in the critical field of health, where
the demand for quality in information is even more critical.
The proposed system is a medical information system that integrates the consumer’s
and the expert’s needs in one application, providing at the same time reliable and
trusted information to the users.
It’s functionality is described using Unified
Modeling Language (UML). Today, UML is accepted by the Object Management
Group (OMG) as the standard for modeling object oriented systems. UML has
evolved from the originally adopted version 1.1 in November 1997 and widely used
for modeling software systems. This thesis uses UML’s latest major revision which is
version 2.0.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
2
1.1 THE SYSTEM
The system is a medical information system that, first of all, fulfills two basic
requirements. First, it provides only reliable and trustworthy sources of
information. Finding high quality medical content on the Internet can be difficult.
The proposed system suggests two ways to get this kind of information. Either by
using existing medical databases such as Medline, or by crawling the Web for sites
accredited by the Health On the Net foundation (called HON accredited sites, see
Glossary). Second, the system can be used effectively both by consumers and experts.
To accomplish this, the system categorizes each document to consumer/expert
categories, with a weight of belief that it belongs to each of these categories. The
benefit of this categorization is that when the user submits a query the system
responds by giving different set of answers depending on the user type, whether a
consumer or expert (e.g. consumers view first the documents that have less medical
terminology so they understand them easily).
The system’s functionality is described using Unified Modeling Language (UML)
which is the state-of-the-art modeling language.
A brief description of this
functionality follows:

The system allows many different ways of indexing a corpus of medical
documents, which in this thesis is called medical document collection (see
Glossary). As mentioned earlier, this collection consists of documents that
belong either to existing medical databases or to Web’s HON accredited sites.

Users must first log in to the system. They find information either by searching
or by browsing. When searching, the user enters a free text query, the query is
mapped to the type of terms that the user has selected before, which might be
either lexicographic(single word terms) or semantic terms (AMTEx, MMTx,
MeSH terms see Glossary) and the retrieval is performed. When browsing the
(expert) user selects the type of document’s categorization, he wants. The
system can support three alternative ways for browsing. The user is allowed to
switch between them and enhance his retrieval.

Also, the system enhances the retrieval performance by combining various
ranking schemes. The documents in the answer set are ordered by a similarity
measure between each document and the user’s query (the classical tf*idf
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
3
weighting scheme), the document’s authority score (for articles only) and the
document’s weight of belief that it belongs to the consumer category (for
consumer users) or to the expert category (for expert users). This ranking
scheme causes the documents that are more likely to belong to the expert
category to rank high in the answer set only if the user is expert. The opposite
happens if the user is consumer. Retrieval performance is also enhanced is by
letting the user activate the query expansion feature. The user can select the
terms used for the expansion from a list of choices such as MeSH terms lower or
higher in the MeSH hierarchy of a query term, synonyms of a query term etc.
Optionally, the expert can filter the documents in the answer set by dropping the
documents that do not belong to his favorite categories.

The retrieval is also enhanced by allowing the users to browse the system’s
database via citation links. This is essential, especially for the expert users, as
they can navigate the literature backward and forward in time.
1.2 CONTRIBUTIONS
It is essential for a system to have a well-designed architecture that is optimized early
and incrementally. Current systems are difficult to expand when the requirements
change or new requirements arise. In practice, most architectural descriptions are
informal documents, usually centered on block diagrams, not supported by tools to
help the programmers.
The goal of this thesis is to describe the system’s requirements of a medical
information system using Unified Modeling Language (UML), the current state-ofthe-art language for system design. UML provides a family of diagrammatic notations
by which a software system can be described at a high level of abstraction and
enforces system modularization, by splitting the system’s functionality into a
collection of connected components. Each component is a replaceable part of the
overall system that fulfils a clear function, evolves independently, can be reused, can
be updated by alternative components. When all of these components have been
implemented they are integrated together to form the complete system.
By using UML the iterative and incremental development of the system is highly
characterized by:

transparency : by decoupling the design from the implementation,
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
4

flexibility : by offering alternative models and implementations, and

extensibility: by allowing
implementation of new models and algorithms
without interfering with existing software e.g. new features are added by adding
new code, rather than by changing the old one.
UML is supported by a variety of tools. In this work, UML diagrams were created
using EclipseUML 2007 Europa for Eclipse 3.3 by Omondo. This tool is selected
mainly because it provides automatic code generator for the UML class diagrams,
thus saving a considerable amount of work for the system’s development stage.
Another benefit of this feature is that later when additional requirements arise (there is
a cycle from requirements to design to code), UML diagrams are also modified, the
automatic code generator is activated again, so there is always a correspondence
between UML diagrams and the implemented system. As a result the system is easily
maintained and enhanced.
Another benefit of the system is that, unlike other existing medical information
systems, it can be used effectively to satisfy both consumer and expert users. The
system allows the user to enter queries in extended length in plain English, when most
other existing systems place limits on query length. This is a great convenience
especially for the consumers, who due to lack of medical knowledge, have difficulties
in choosing accurate medical phrases as a starting point for their search. Also, the
system supports browsing the database via citation links. This feature already exists in
other fields (e.g. CiteSeer in the computer science field etc), but not in the medical
domain field.
1.3 THESIS STRUCTURE
The remainder of the thesis is organized as follows: First (chapter 2), some basic facts
for both of the thesis fields (modeling and medical information systems), are
introduced. The concept of modeling, the diagrams of UML 2.0, the four diagrams
that are of special importance for this thesis, the use cases tool are explained in brief,
followed by a description of tools (MMTx and AMTEx) that automatically map free
text into medical terms and are used for indexing and retrieval by the system. Finally,
existing medical information systems (e.g. digital libraries, medical search engines,
portals etc) are explored and briefly presented.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
5
After the basics (in chapter 3) the system’s architecture is presented, followed by a
detailed description of the system’s functional requirements based on use cases and
the four UML diagrams (package, use case, activity and class diagrams).
The last chapter (chapter 4) presents the part of the system implemented so far.
Finally a Glossary of the terms used in the requirements analysis phase is attached
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
6
Chapter 2
BACKGROUND & RELATED WORK
In this chapter, UML, which is the current state-of-the-art system modeling language,
is presented. From UML only the diagrams used for the system’s modeling are further
described. These are the package, use case, activity and the class diagrams. Then the
use cases tool, used for capturing the system’s functional requirements, is presented.
The system maps both documents and queries to terms from MeSH and UMLS
terminology, which are used for indexing and retrieval, so a brief description of these
vocabularies follows. Then, the tools used for this mapping (U.S. National Library of
Medicine MMTx and AMTEx) are also highlighted.
Finally, CiteSeer, a scientific literature digital library and several medical information
systems (medical search engines, portals, medical meta search engines etc) that
already exist in the Web are briefly presented.
2.1 UML (Unified Modeling Language)
Modeling, the design of a system before coding, is an essential part for every
application development. System modeling is considered to be successful when the
following requirements are met: the system’s functionality is formally described and
correct, the end-user needs are met, scalability and extensibility are supported, the
system’s design is visualized and checked before the coding starts, etc. UML is the
current state-of-the-art language. It must be mentioned that UML is not just a list of
diagrams, it is a knowledge representation language. This means that each
diagram’s element, e.g. box, line, rectangle, etc, is supported by certain syntax and
semantics.
UML was selected for the system’s modeling for several reasons. First of all it is an
industry standard, controlled by the Object Management Group (OMG) an
international, open membership, not-for-profit consortium of companies. Another
benefit is that UML is built upon fundamental Object Oriented concepts (e.g. class
and operation), so it helps better in describing, designing and developing particularly
object-oriented software. Also, as mentioned earlier (see Contributions) UML is
supported by a variety of tools that not only help the designer to ‘draw’ the diagrams,
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
7
they also offer advanced features, such as building a UML diagram from existing
code in order to understand it (reverse-engineering) or generating program language
code from a UML diagram (forward engineering). Also, it is easy to transfer a UML
model from one tool into another by using the XMI (XML Metadata Interchange,
which is another OMG standard). Other benefits in using UML is that it emphasizes
Use Case Analysis, which is a valuable tool for capturing the process from a user’s
perspective (see section 2.2). UML also allows the designer to select the level of
detail he wants in his description.
Finally, UML makes the system’s design modular and easily extensible. Plugins can
be added to alter the process at every stage of the system’s architecture, e.g. in parsing
and analyzing each individual document from the medical document collection, in
categorizing each document in three different types of categories, in extracting
different type of terms from both documents and queries, in writing the index data
structures, in retrieval, in browsing, etc. As a result, the system is easily maintained
and enhanced later even when the initial system’s programmers or requirements have
changed. This is essential as it is often more difficult to maintain a system than
building it from the beginning.
UML 2.0 defines thirteen types of diagrams, divided into two categories:

Structure Diagrams show the static structure, the ‘things’ that make up the
system being modeled (e.g. the classes, the objects, the relationships among
them, etc), irrespective of time. Structure diagrams are the following : Class
Diagram, Object Diagram, Component Diagram, Composite Structure Diagram,
Package Diagram and Deployment Diagram

Behavior Diagrams emphasize what must happen in the system being modeled,
such as the interactions with the users, the various states of the system as it
executes over time, the effects and the results of an operation or an event etc.
Behavior diagrams are the following: Use Case, Activity Diagram and State
Machine Diagram and the Interaction Diagrams, which include the Sequence
Diagram, Communication Diagram, Timing Diagram, and the Interaction
Overview Diagram.
Each of the diagrams mentioned above views the system from a different perspective,
e.g. the use case diagrams show how users interact with the system, the class
diagrams show the system’s classes and their relationships, the activity diagrams
show the details of how a use case works, the package diagrams group classes into
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
8
packages when the classes of the application become too many, the sequence
diagrams describe also a use case (like the activity diagrams), but it shows the
objects and the messages that are passed between these objects etc.
Four diagrams are found to be appropriate for the system’s modeling. A brief
description for each one of them follows (the symbols and notation used in these
diagrams are explained briefly in the corresponding sections):
1) Package diagram (see section 3.1): Package diagram is used only when the
system is large enough, e.g. many use cases, classes etc. and the model must be
divided into ‘packages’. Package diagrams show these packages and the interactions
between them at a high level. Each UML package organizes the model’s use cases,
classes, or both into groups.
2) Use case diagram (see section 3.2): The highest level is the use case diagram
which captures the behavior of a system as it appears to an outside user. It describes
all of the system’s functionality in a graphical table of contents (what the system
should do, not how it should be done). The use case diagram has four elements: (a)
the actors, who are persons, an organization, or external systems (in this work the
actors are only persons), etc. Each actor is associated with several use cases, each of
which describes what the actor wants to do with the system (this is the actor’s goal),
for instance ‘retrieve documents in respect to a query’ and the user interacts with the
system in order to achieve this goal, (b) the use cases (represented with ovals with
their names on it), that describe a sequence of actions that must be executed in order
to achieve or abandon (e.g. the user can’t register) the actor’s goal, (c) the lines that
represent several types of relationships (e.g. isa, include, extend etc) between an
actor and a use case, between two use cases and between two actors and (d) the
system boundary which is a rectangle that includes all the use cases which indicate
all the system’s functionality.
3) Activity diagrams(see section 3.4): Activity diagrams are useful for analyzing a
use case by describing what actions need to take place (the use case scenarios see
section 2.2) from the system’s point of view. Strictly, an activity is a sequence of
actions. The main reason to use them is that they model the workflow behind the
system being designed. For each use case they show a sequence of steps (actions) and
the flow of control from one step to another. These steps can cover multiple use cases.
There are many different ways to use them. Here activity diagrams concentrate
mostly on ‘what gets done’ in a use case scenario, rather than on ‘who does each
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
9
action’ (which class is responsible for each action).
Activities diagrams are
recommended when the system has parallel activities (multiple activities that occur at
the same time) and identifying these activities early in the requirement analysis phase
is essential.
4) Class diagrams (see section 3.5): Class diagrams are widely used. They describe
the structure of the system e.g. the types of objects, the various kinds of relationships
that exist among them, the properties and operations of each class etc. From these
diagrams, there are tools that automatically generate code in a programming language
(such as Java) and this code helps the team work on the same set of files without
interfering with each other’s work.
2.2 USE CASES
A use case is a collection of possible scenarios between the system and the external
actors. Each use case is characterized by the goal that the primary actor has. This goal
might be delivered or fail.
They are a widely used, valuable tool for capturing the functional requirements of a
system. These requirements can be expressed either in text or (better) in a more
structured format to avoid inconsistencies and redundancy. Here, the one-column
table template is used [5] :
UC-x
Use case title (usually it is an active verb phrase that names the goal
of the primary actor e.g. ‘Retrieve relevant documents’, where the phrase
starts with the active verb ‘Retrieve’)
A brief description of the use case’s behaviour is (use case brief) is presented.
Goal in Context
A longer statement, which describes the use case’s goal in more detail.
Details
Primary Actors: Are those with the goal that the use case addresses
Preconditions: what is already true, before the use case starts. Since it is known that
this condition is true, it will not be checked again during the use case (e.g. for the use
case ‘Retrieve relevant documents’, the precondition is that ‘the user has entered a
non-empty natural language query’).
Level: the goal level is one of the following (note that the user goal use cases must be
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
10
detected first because these are the important ones and justify the system’s existence):
-
Summary: A summary level goal use case provides a table of contents for the
lower-level use cases (which are user and other summary goal use cases)
-
User: A user level goal use case provides the goal the primary actor has when
using the system. The lists of these use cases make most of the system’s
functionality.
-
Subfunction : A subfunction level goal use case carries out the user goals
mentioned above. They are usually needed because many other user goals use
cases, use them.
Scope : each use case is labelled with one of the following :
-
System : only the first use case that glues together all the others has this scope. It
includes all the system’s functionality.
-
The appropriate subsystem’s name (see the system’s architecture shown in Figure
3.3). That means that the main system is opened and the use case describes how a
piece of it works e.g. the System Management subsystem, the Document Retrieval
subsystem etc.
Triggering Action:
It states what event gets the use case started. Here, the
triggering action is the first step of the use case.
Flow of Events
Main Success Scenario (is the body of the use case):
The main success scenario describes the case where nothing goes wrong. It is a
sequence of numbered steps (a well written use case has more that 3 and less than 9)
where each step is a simple statement and clearly shows who is controlling the action
(who has the ball). It must be mentioned that here only active verbs that move the
process forward are used e.g. the verb check is not used because it leaves open what
the result of the check (either passes or fails). Instead the verbs ensure, verify and
other goal achieving verbs are used. A typical example is instead of writing : ‘The
system checks for the type of user and …..’the following is written: ‘The system
verifies that the user is a consumer and ….’ or ‘The system verifies that the user is an
expert and ….’
Also it must be mentioned that when a use case references another use case, the
second one is written in blue and underlined (like a hyperlink).
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
11
Flow of Events
Extensions:
The Extensions describe what can happen differently during the main success
scenario. Extensions can lead to a success (the actor achieves the use case’s goal) or
to failures.
Each extension consists of two parts:
(a) the extension conditions which are the conditions under which the system takes a
different behaviour. A colon (:) is put after the conditions.
(b) the extension handling which is a sequence of steps to deal with the conditions.
Variations:
Variations to a step from the main success scenario exist because what is happening in
a step is the same, but there could be several different ways to do this. For example,
see UC-10 Categorize the document to consumer/expert category. Here there is a step
(step 2) that categorizes the document as an expert one if it has at least one expert
term. There is also a variation to this step that categorizes the document as an expert if
it contains at least 2 expert terms. Notice that in both cases the categorization will be
done, what is different is how this is done.
2.3 MeSH
MeSH1
(Medical Subject Headings) is a taxonomic hierarchy of medical and
biomedical terms suggested by the U.S. National Library of Medicine2 (NLM). It
contains almost 25,000 terms covering life sciences, medicine etc. MeSH vocabulary
reflects the progress in biomedical sciences. Every year several hundred new terms
are added to the MeSH vocabulary and some existing terms are modified. MeSH is
used by numerous health organizations and medical libraries, including MEDLINE,
to organize materials and index information.
The MeSH vocabulary is available in XML format [8]. MeSH headings (MH) are
organized in a tree with 16 main braches. Each branch has many levels of subbranches, and each heading has a position in the hierarchy which is called tree
1
2
http://www.nlm.nih.gov/mesh
http://www.nlm.nih.gov
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
12
number. Some terms appear in more than one branch of the tree, e.g. the term ‘ear’
appears in two locations in the hierarchy, the ‘Body Regions – Head’ and the ‘Sense
Organs’, therefore it carries two different tree number, [A01.456.313] and [A09.246].
Most of the MeSH heading are accompanied by a list of synonyms or very similar
terms, known as entry terms. They indicate that information related to one term will
be found under a different term, e.g. ‘nose bleed’ and ‘Nosebleed’ are entry terms for
the MeSH heading ‘epistaxis’. Also each MeSH heading is associated with the scope
note which is a short piece of free text that defines the corresponding MH.
2.4 UMLS
UMLS3 (Unified Medical Language System) includes many controlled vocabularies
in the biomedical sciences, designed and maintained by the US National Library of
Medicine. It consists of the following components:

Metathesaurus, which is a very large multi-purpose and multi-lingual
vocabulary that contains information about biomedical and health related
concepts, their various names and the relationships among them. It is designed
for use by system developers and built from the electronic versions of over 100
sources, e.g. thesauri, vocabularies etc called “source vocabularies” of the
Metathesaurus. Some of them are Dental Terminology, the Diseases database,
Gene Ontology, International Classification of Primary Care, MeSH etc

Semantic Network, which consists of the semantic types that provide a
consistent
categorization
of
all
concepts
represented
in
the
UMLS
Metathesaurus, and the semantic relations that exist between semantic types.
The primary link between the semantic types is the ‘isa’ link. The 'isa' link
establishes the hierarchy of types within the Network and is used for deciding on
the most specific semantic type available for assignment to a Metathesaurus
concept. There is also a set of non-hierarchical relationships, which are grouped
into five major categories: `physically related to,' `spatially related to,'
`temporally related to,' `functionally related to,' and `conceptually related to.'
[9]
3
http://umlsks.nlm.nih.gov
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
13

SPECIALIST Lexicon which is intended to be a general English lexicon that
includes both commonly occurring English words and biomedical vocabulary. It
is used for natural language processing.
2.5 The MMTx tool
MMTx4 (MetaMap Transfer) is a tool from the U.S. National Library of Medicine,
that maps biomedical text (from documents, queries) into concepts in the UMLS
Metathesaurus. MMTx uses the Metathesaurus and the SPECIALIST lexicon during
the term extraction process. This process works in the following steps [3]: First, the
text is parsed and a simple linguistic filter isolates the noun phrases. Then variants are
generated from the resulting phrases, using the SPECIALIST lexicon, e.g. the phrase
‘obstructive sleep apnea‘ has variants: obstructive sleep apnea, sleep apnea, sleep,
apnea, osa, etc. Afterwards, the candidate concepts from the UMLS Metathesaurus
are retrieved and evaluated against the phrases, e.g. for the variant osa the following
Metathesaurus concepts are retrieved: osa [osa antigen],osa [osa gene product], osa
[osa protein] etc. Finally, the best of the above candidate concepts are chosen.
MMTx has some limitations especially during the variant generation stage [3] where
the expansion of the initial text phrase to all possible variants is quite exhaustive. As
a result, this process also results in term over-generation, which diffuses the original
term concept and unrelated terms are added to the final candidate list. Another
limitations is that MMTx extracts general UMLS Metathesaurus terms rather than
MeSH (but MEDLINE indexing is based on MeSH).
2.6 The AMTEx method
AMTEx (Automatic MeSH Term Extraction method) is an automatic term extraction
method, specifically designed for the automatic indexing of documents in large
medical collections such as MEDLINE. AMTEx first uses the C/NC-value method
for term extraction (C/NC value is domain-independent and combines statistical and
linguistic information for the extraction of multi-word and nested terms, that are very
common in the biomedical domain). Finally it validates the extracted terms against
MeSH, rather than the full UMLS Metathesaurus (that contains over a hundred source
vocabularies, including MeSH see section 2.4) mapping of MMTx [15].
4
http://mmtx.nlm.nih.gov
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
14
2.7 RELATED SYSTEMS
The amount of health data accessible on the Web is increasing and Internet has
become a major source of health information. Many medical information systems
such as search engines, portals, meta search engines, digital libraries etc are available.
Here is a brief description for some of these existing systems:
Experts are interested (and this feature is not supported by the existing systems) in
browsing the medical literature via citation links, which means that it’s important for
them to navigate backward (the list of cited articles) and forward in time (the list of
subsequent articles that cite the current article).There are systems e.g. CiteSeer5 that
support this feature but not in the medical domain. CiteSeer allows navigation in the
computer science literature. It downloads articles (e.g. Postscript files) that are made
available on the Web, converts them to text and performs Autonomous Citation
Indexing (ACI). During this process the article’s citations are parsed, citations to the
same article but in different formats are identified, the context of citations within the
body of the articles are also identified etc. Finally, CiteSeer performs full-text
indexing of the articles and citations as well [13]. It can also generate statistics on
citation frequency and allows the estimation of the importance of articles. CiteSeer’s
disadvantage is that it has grown enough pushing the limits of the current system’s
capabilities; it is hard to administer and modify. To overcome these drawbacks
CiteSeerχ is introduced [14].
Medscape6 is used only by experts and provides clinical information relevant to
their chosen speciality. Also, experts are interested in searching bibliographic
databases e.g. MEDLINE through PubMed. MEDLINE is a database of over 15
million medical and scientific articles (most of them in English) indexed from the
sixties to date, created and maintained by the U.S. National Library of Medicine
(NLM). PubMed7
is developed and maintained by the National Center for
Biotechnology Information (NCBI) at NLM. It provides access to MEDLINE and to
articles in selected life sciences journals not included in MEDLINE through an easy
to use free Internet site. MEDLINE indexers describe the content of biomedical
articles by assigning (typically 10 to 12 per article) the most specific MeSH terms(see
section 2.3). PubMed uses them for retrieval and the search strategy is enhanced (e.g.
5
http://citeseer.ist.psu.edu
http://www.medscape.com
7
http://www.ncbi.nlm.nih.gov/PubMed/
6
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
15
the query ‘bad breath’ is mapped automatically to the MeSH term ‘halitosis’). Within
PubMed’s Consumer Health link leads to MedlinePlus8 which is intended to be used
mostly by consumers and also provides information that is authoritative and up to
date.
Other medical information systems that are used mainly by consumers are the
following. MedicineNet.com9, provides in an easy to understand language,
authoritative medical information only for consumers. These are articles are written
by a network of health professionals and that’s the system’s drawback since it’s
content is not always up to date. Health and medical information can also be provided
by portals with too confusing user interface, such as
WebMD10 ,MEDNETS11
Healthline12 etc, by metasearch engines such as OmniMedicalSearch.com13 that
also offers a special link for the experts.
Also there exist medical search engines used by both experts and consumers. Some
of them are listed below: MedHunt14, developed and maintained by the Health On
the Net Foundation (HON - a not-for-profit organization that aims to provide access
to reliable sources of online medical information). MedHunt retrieves medical
information either from HON’s accredited sites or from medical pages crawled from
the Web. WRAPIN15 (Worldwide online Reliable Advice to Patients and Individuals)
developed from the EU-WRAPIN project, that also uses medical trustworthy sources
such as the NLM’s PubMed, HON’s MedHunt, U.S. Food and Drug Administration
(FDA) etc. WRAPIN also supports different types of query from a few keywords to
entire web pages (specified by their URL), it maps both query and documents to
MeSH terms(the HONMeSHMapper module is used) that are used for indexing and
retrieval.
The proposed system selects some interesting features of the above systems(such as it
uses documents only from authoritative sources to create the system’s database,
allows the users to navigate though literature via citation links etc) and adds more
functionality (such as it improves ranking by
8
http://medlineplus.gov
http://www.medicinenet.com
10
http://www.webmd.com
11
http://www.mednets.com
12
http://www.healthline.com
13
http://www.omnimedicalsearch.com
14
http://www.hon.ch/MedHunt
15
http://www.wrapin.org/
9
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
the article’s authority score, the
16
document’s consumer/expert score etc).in order to be efficient and effective for both
consumers and experts. See section 1.1. and section 1.2 for more details about the
system’s features.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
17
Chapter 3
REQUIREMENTS ANALYSIS
During the analysis phase the system’s functionality is defined. The techniques used
to capture this functionality are described in sections 2.1 and 2.2. During this phase is
also important to discover classes, associations between classes, class operations, etc.
The results of the analysis (UML diagrams and use cases) serve as input to the next
phase.
First the package diagram (see section 3.1) is presented which shows the system’s
packages and their relationships. Then the use case diagram (see section 3.2) follows
and highlights the relationships between the three different type of users
(administrator, expert, consumer) and the system’s use cases. For simplicity, only
summary and user goal use cases are presented. Then the seventeen use cases (see
section 3.3) are presented. For each use case an activity diagrams (see section 3.4) is
created, in order to show graphically, from the system’s point of view, each use case’s
behavior in more detail. Finally, the UML’s backbone, the class diagrams (see section
3.5) are presented. There are four class diagrams, one for each package described in
the package diagram.
3.1 PACKAGE DIAGRAM
Initially there was no need to create a package diagram. But as the system’s
functionality increased and the system became larger, more classes were introduced
leading to a very complicated class diagram. Then it was decided to split classes (and
use cases as well) into packages and the package diagram was created. The system’s
package diagram is shown in Figure 3.1. The packages are displayed with a tabbed
folder with the package’s name written on it. Also it shows the dependencies between
the packages. There is a dependency from one package to another if any class in the
first package has a dependency to any class in the other package.
Each package leads to a UML class diagram (see section 3.5), and each class is a
member of a single package. In Java terms UML packages correspond to java
packages (every class in the package must have a unique name within its owning
package). It must be mentioned that the package diagram, was very helpful to control
the large-scale structure of the proposed system.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
18
Here follows a brief description for each package:
The User_Interface_Package contains the classes that interact with the user for query
entry,
results
presentation,
user
login,
user
registration
etc.
The
Document_Retrieval_Package contains the classes that are mainly responsible for
processing the user’s query, fetch the retrieved documents (it interacts with the
System_Management_Package) and display the results to the user (it interacts with
the User_Interface_Package). The User_Management_Package contains the classes
that manage the users. Finally, the System_Management_Package contains the entity
classes (these are classes that include the data that must be stored in the system’s
database).
Figure 3.1: Package diagram
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
19
3.2 USE CASE DIAGRAM
The use case diagram is like a graphical table of contents that presents the system’s
behavior in respect to the users. The system’s use case diagram (see Figure 3.2)
presents:
 the system’s actors. Here the actors (expert, consumer and administrator) are
only humans. See UC-1, Use the application for a brief description for each
actor’s profile (e.g. job, typical background, skills etc).
 a list of use cases (represented as ovals). In order to keep use case diagram as
clear as possible, low level system interactions (that is use cases lower than user
goal level) are not presented.
 relationships (represented as arrows). Two relationship types are presented.
Those that show the inheritance relationship between the actors (the expert user
is a consumer plus some extra functionality that the system offers to him e.g.
browsing the system’s database by medical topics). The other is the <<include>>
relationship, a directed relationship between two use cases. It denotes the
invocation of a use case by another one, e.g. the behavior of the use case ‘Present
results to user (UC-15)’ is inserted into the behavior of the ‘Answer the query
(UC-12)’ use case.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
20
Figure 3.2 : Use case diagram
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
21
3.3 USE CASES
The system’s architecture is shown in Figure 3.3 below. The 17 use cases are split up
into the following modules:

System Management,

Document Retrieval,

User Management, and

User Interface Module
Medical
document
collection
System Management Module
Document
Analysis
Indexing
System’s
database
Document Retrieval Module
User Management Module
Query
Handling
Browsing
results
User Interface Module
Figure 3.3: The system’s architecture
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
22
The use cases have a hierarchical structure (see Figure 3.4) and each use case
belongs to the module(s) mentioned in Figure 3.3. To be more specific the use cases
written in blue belong to the User Interface Module, in orange to the User
Management, in green to the Retrieval and in red to the System Management Module.
Each use case belongs to exactly one module except from
UC-9, Create the
representation vectors for the document or the user's query, which belongs to two
modules, the System Management Module and the Retrieval Module (it is both red
and green).
(UC-1) Use the application
(UC-2) User
registration
(UC-3) User
login
(UC-5) Create the
system’s
database
(UC-7) Build the
citation indexes
(UC-4) System
management
(UC-6) Update
the system's
database
(UC-8) Extract
the document's
terms and
categorize the
document
(UC-9) Create the
representation vectors
for the document or
the user's query
(UC-12) Answer
the query
(UC-13) Retrieve
relevant documents
(UC-11) Build indexes for
the document's terms
(UC-10) Categorize
document to
consumer/expert
category
(UC-17) Browse
database by medical
topics
(UC-15) Present
results to user
(UC-16)
Reformulate
the user's query
(UC-14) Rank the
documents
Figure 3.4 :Usecase’s hierarchical
structure
(Α) System Management Module
(B) Retrieval Module
(C) User Management Module
(D) User Interface Module
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
23
A brief description for each module and a list of the corresponding use cases is
followed.
(A) System Management Module
This module parses each document from the medical document collection (see
Glossary) and adds them to the system’s database (see Glossary).
It includes the following use cases:

System management (UC-4), where the system's database is created/updated

Create the system’s database (UC-5), where the system parses each
document from the medical document collection, analyzes and indexes its
content. Also, it categorizes the document and assigns an authority score for
the articles only.

Update the system’s database (UC-6), where the system processes the
updated or the new documents from the medical document collection and
updates the system's database accordingly.
Before adding the document to the system’s database, it must be processed by the
following subsystems (see Figure 3.3):
(I) The Document Analysis subsystem which parses the document, extracts it’s
(semantic and lexicographic) terms and categorizes it to Semantic Network MeSH
categories (see Glossary) and to consumer/expert categories. The Document Analysis
subsystem includes the following use cases:

Extract the document's terms and categorize the document (UC-8), where
the system extracts the document's terms and creates three representation
vectors. Finally, the system categorizes the document to: (a) consumer/expert
categories, and (b) Semantic Network MeSH categories.

Create the representation vectors for the document or the user's query
(UC-9), where the system extracts terms from the document and creates three
representation vectors.

Categorize the document to consumer/expert category (UC-10), where the
system categorizes the document as consumer or expert with a weight of
belief that it belongs to each category
(II) The Indexing subsystem that builds the citation indexes and the document’s
terms indexes (extracted from the Document Analysis subsystem described above).
The Indexing subsystem includes the following use cases:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
24

Build the citation indexes (UC-7), where the system processes the
bibliography part (only for research documents) and builds the citation
indexes.

Build indexes for the document's terms (UC-11), where the system indexes
the document by using the Inverted Files method.
(B) Retrieval Module
This module retrieves documents from the system’s database either in respect to the
user’s query or by browsing the medical topics. It’s obvious that this module consists
of the following subsystems (see Figure 3.3):
(I) The Query Handling subsystem which parses the user’s query, extracts it’s
(semantic and lexicographic terms), retrieves and ranks a list of relevant documents to
the query and finally it suggests a list of terms for the query expansion process, if it is
necessary. The Query Handling subsystem includes the following use cases:

Retrieve relevant documents (UC-13), where the system compares the user's
query with each document from the database and retrieves a list of relevant
documents in response to the user’s query.

Create the representation vectors for the document or the user's query
(UC-9), where the system extracts terms from the user’s query and creates
three representation vectors. Notice that this use case also belongs to the
Document Analysis subsystem from the System Management Module described
above.

Rank the documents (UC-14), where the system uses one of the: (a)
lexicographic terms or (b) MMTx terms or (c) AMTEx terms representation
vectors for both the query and the documents and produces a ranked list of
retrieved documents.
(II) The Browsing subsystem. Only one use case belongs to this subsystem:

Browse database by medical topics (UC-17), where the system allows the
Expert user to view the documents through browsing the system’s database by
the medical topics.
(C) User Management Module
This module manages the user’s database. It includes the following use cases:

User registration (UC-2), where the user registers to the system.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
25

User login (UC-3), where the system validates the user (administrator,
consumer or expert). In case where he hasn’t registered, the system forces him
to register.
(D) User Interface Module
This module provides the system’s functionality to its users. It includes the following
use cases:

Use the application (UC-1), where the user uses the system's functionality
according to his type (Administrator, expert, consumer, see UC-1 about each
user’s profile))

Answer the query (UC-12), where the user enters a query, retrieves relevant
documents and reformulates the query, if the results are inefficient.

Present results to user (UC-15)

Reformulate the user's query (UC-16), where the user's query is expanded
with new terms.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
26
UC-1
Use the application
This is the higher level use case and glues together the lower level ones. It is little
more than a table of contents that names the actors and their highest level goals. In
our system all actors are humans. It is important to refer not just to the actor’s names
but also to each actor's profile (brief description that includes information such as
their job, typical background, skills etc). This kind of information is useful later in the
system's behavior and the user's interface design so the software will suit the needs of
the end users.
Our system supports three different types of actors:
1. the Administrator, who is a computer science expert. He creates the database,
manages the application's parameters and he is not expected to be familiar with
the medical terminology.
2. the Consumer, who is an ordinary user (patient, relative, the public generally).
He is usually unfamiliar with the medical terminology, having a low degree of
health literacy. He is often unclear about the problem that he is facing. As a result,
it is difficult for him to choose a few accurate medical phrases as a starting point
of his search. He is also not expected to operate a GUI with ease and may use the
system occasionally.
3. the Expert who is a health care professional (medical doctor, nurse,
physiotherapist, nutritionist, medical researcher etc), therefore familiar with the
medical terminology. He might have difficulties in operating a GUI, he might
want to customize the User Interface and he is expected to use the system
frequently.
The actors mentioned above, use the system in order to accomplish some goals. The
highest level goals for each actor are presented in the use case's Main Success
Scenario steps shown below:
Goal in Context
The user uses the system's functionality according to his type (Administrator, Expert,
Consumer)
Details
Primary Actors: Administrator, Consumer, Expert
Supporting Actors:
Preconditions: none
Level: Summary
Scope : System
The user opens the application
Triggering Action :
Flow of Events
Main Success Scenario:
1. The system provides the User with the following list of options and the user logs
in :
1.1. Register to the system (UC-2)
1.2. Login to the system (UC-3)
2. The system verifies that the user is the Administrator and provides him the option
to manage the system (UC-4)
3. The system verifies that the user is an Expert or a Consumer and provides him
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
27
Flow of Events
with the following options:
3.1. The user enters a query, the system retrieves relevant documents in response
to his query and the user reformulates the query if necessary (UC-12)
3.2. The user customizes his profile from the following list of (advanced) options:
3.2.1. update the number of the top documents from the retrieved list. MeSH
terms will be extracted from these documents and will be used for the
query reformulation process
3.2.2. update the similarity threshold (it defines terms lower or higher in the
MeSH hierarchy used for the query reformulation process)
3.3. The user updates his personal information, username and password
4. The system verifies that the user is an expert one and allows him to browse the
documents through the Semantic Network MeSH categories (UC-17)
5. The system repeats step 2-4 until the user indicates that he wants to exit the
application
6. The system terminates the session
Extensions:
1.a The user’s registration or the user’s login process fails:
1. The system continues to step 1 until the user succeeds to login or to register
3.3.a The user updates his username:
1. The system verifies that the username has not been used by another user
3.3.b The user is an expert one:
1. The system allows him to update his favorite medical topics
Variations:
4. The expert can switch to browse to another type of document's categorization such
as: (a) categories that come from the document clustering process or (b) categories
that come from the citation graph Link Clustering process.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
28
UC-2
User registration
Only consumer and expert users can register to the system. They provide necessary
information for the registration such as personal information, username, password etc.
Additionally, the experts can select their favorite medical topics from the Semantic
Network MeSH categories (see Glossary). Alternatively, the they can select their
favorite medical topics from categories that are created through (a) the document
clustering process or (b) the citation graph Link Clustering process.
Goal in Context
The user provides information to the system and the system registers him as a valid
user
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: The user was not a valid user
Level: User
Scope : User Management Module
The user registers to the application
Triggering Action :
Flow of Events
Main Success Scenario:
1. The user enters personal information such as his name, last name, email, job,
phone number(s), addresses(s) and the type of user he is(consumer or expert)
2. The user enters a username and a password
3. The system verifies that the username has not been used by another user
4. The system stores the user's personal information, username, password and the
registration date (which is the current date)
5. The system verifies that the user is an expert and allows him to select favorite
medical topics from the Semantic Network MeSH categories
Extensions:
3.a The username has been used by another user:
1. The user is prompted to enter a new username
2. The system continues to step 3
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
29
UC-3
User login
The system validates the user (administrator, consumer or expert). In case where the
user hasn’t registered, the system forces him to register (see UC-2 User registration).
Goal in Context
The user enters the application
Details
Primary Actors: Administrator, Consumer, Expert
Supporting Actors:
Preconditions: The user is the system's Administrator or he is a registered user
(Expert or Consumer)
Level: Subfunction
Scope : User Management Module
The system asks the user to provide his username and
Triggering Action :
password
Flow of Events
Main Success Scenario:
1. The user enters his username and password
2. The system validates the username
3. The system validates the password
Extensions:
2.a Invalid username:
1. The user registers in the application (UC-2)
3.a Invalid password:
1. The system notifies user and requests the password again
2. The user re-enters the password
3. The system continues to step 3
3.b Invalid password entered too many times:
1. The system notifies the user
2. The system terminates the session
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
30
UC-4
System management
This is a high level use case, a table of contents, that names the system
Administrator’s goals which are :
(1) He defines the medical document collection (see Glossary). This collection is
created either:
a. by using an existing medical database such as Medline (see Glossary)
or
b. by crawling the Web's HON accredited sites (see Glossary). The
crawling process starts with a small set of HON accredited sites (the
'seed') and then explores other pages by following the page's links.
These sites are identified easily by the a HON-icon in the home page
or in the first level pages (pages that are linked from the home page).
This HON-icon is associated with a link to the site's HON certificate,
which includes details about the HON's accreditation process such as
the date of initial review, the HONcode PIN number, the date of last
subsequent review etc.
(2) Once the medical document collection is defined, the administrator's main
goal is satisfied, which is to create the system’s database (see Glossary). The
documents from the medical document collection are parsed one by one,
analyzed and indexed (see UC-5 Create the system’s database).
(3) Also, one of the system administrator's goals is to customize the system's
parameters (see step 1.3 for details) such as how often the re-crawling process
(if the medical document collection is created by crawling) is triggered, e.g.
once a month, every two months etc. When the re-crawling process finishes
the system's database is updated (see UC-6 Update the system’s database).
Goal in Context
The system manages the system's database and customizes the system's parameters
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The system's Administrator is logged on to the system.
Level: Summary
Scope : System Management Module
The system presents a list of options to the Administrator
Triggering Action :
Flow of Events
Main Success Scenario:
1. The Administrator views the following list of options and selects one of them:
1.1. Create the system’s database:
1.1.1. Create the medical document collection either:
1.1.1.1.
by using existing medical databases, or
1.1.1.2.
by crawling the Web's HON-accredited documents
1.1.2. The system parses each document from the medical document
collection one by one, and creates the system's database (UC-5)
1.2. Update the system’s database:
1.2.1. Update the medical document collection either:
1.2.1.1.
by adding/updating articles from latest versions of existing
medical databases, or
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
31
Flow of Events
1.2.1.2.
by re-crawling the Web's HON-accredited documents
1.2.2. The system updates the system's database (UC-6)
1.3. Customize the system’s parameters from the following list:
1.3.1. the crawling parameters such as the crawling strategy, the seed URL's,
how often the re-crawling process starts etc
1.3.2. parameters that are used for the query expansion process such as the
number of the top-ranked retrieved documents, the similarity threshold
etc
1.3.3. define the type of the medical categories that expert users use for
browsing the medical documents (default are the Semantic Network
MeSH categories)
1.4. The Administrator selects from the following list of options:
1.4.1. he defines a period of time, the system lists information about the users
that have been (or have not been) using the system during this period
1.4.2. he notifies users with new articles, medical topics etc
Extensions:
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
32
UC-5
Create the system’s database
This use case is mainly responsible for managing the document indexing and
categorization process.
Prior to indexing, the system iterates through the documents of the medical document
collection and for each individual document (such as html, plain text documents,
Microsoft Word, Excel, PowerPoint, Adobe Acrobat files etc):
(1) It parses and extracts the document’s parts (all parts are optional). These are
the document’s
 title,
 text,
 bibliography part (for articles only),
 metadata, such as information about the authors(name, lastname etc)
and their affiliation(university, departure name etc), the article's source
e.g. journal, book, conference etc, the date of the article's publication
or the date of the page's creation and finally the Medline terms (only
for the Medline articles)
(2) It builds the author and the Medline terms indexes (if they exist)
(3) It processes the bibliography part (only for research documents) and builds
the citation indexes (see UC-7 Build the citation indexes). From now on the
terms article, paper and research document will be used interchangeably.
(4) It extracts the document's terms(semantic or lexicographic) and creates the
corresponding document representation vectors(see UC-9 Create the
representation vectors for the document ot the user’s query).Then, based on
the document's MeSH terms representation vector, the system categorizes the
document to one of the consumer/expert categories and to Semantic Network
MeSH categories (see UC-8 Extract the document's terms and categorize the
document).
(5) It creates (optionally) another representation vector for the document which
consists of user-defined MeSH terms. At this point, the Administrator uses the
system on behalf of a medical expert.
Once the representation vectors mentioned above are created, they are passed to the
Indexer, which creates four main data structures (see UC-11 Build indexes for the
document's terms ).
When all the documents from the medical document collection are indexed and stored
to the system's database(or the system’s resources are finished), the system computes:
 each term's frequency in the whole collection and stores that value in
the corresponding term's Vocabulary and
 an authority score based on the citation graph link analysis process.
The system uses these values later on the retrieval process (see UC-13 Retrieve
relevant documents).
Finally, further document categorization can be performed through the clustering
process based either on:
 the citation graph link analysis or
 document clustering using the document vector representations (e.g. by
kMeans, xMeans etc).
Goal in Context
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
33
The system parses each document from the medical document collection, analyzes,
indexes its content, categorizes it and assigns an authority score for the articles only.
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The medical document collection has been created
Level: Summary
Scope : System Management Module
The system gets a document from the medical document
Triggering Action :
collection
Flow of Events
Main Success Scenario:
1. The system converts the document into text and extracts it's structural parts
2. The system builds the index for the authors field
3. The system ensures that the document is a research one by testing for the
existence of a bibliography part and builds the citation indexes (UC-7)
4. The system extracts the document's terms, creates the corresponding
representation vectors and categorizes it to : (a) consumer - expert categories and
(b) Semantic Network MeSH categories (UC-8)
5. The administrator assigns user-defined MeSH terms to the document
6. The system builds the indexes for the document's terms (lexicographic, MeSH,
UMLS) (UC-11)
7. The system gets the next document from the medical document collection and
repeats steps 1-6 until all of them are stored in the system's database
8. The system augments the vocabularies of the database by the addition of global
statistics about each term that occurs in the document collection
9. The system performs link analysis to the citation graph, computes and stores an
authority score for each research document
10. The system performs clustering based on the citation graph and achieves further
categorization for the research documents
11. The system performs document clustering and extracts further document
categorization
Extensions:
5.a The administrator doesn't assign user-defined MeSH terms to the document:
1. The system continues to step 6
7.a The system resources are finished:
1. The system continues to step 8
Variations:
1.a. For the ps or pdf files the conversion into text can be accomplished using the
PreScript from the New Zealand Digital Library project
1.b. The document's parts might be extracted using various models, such as regular
expressions, classifiers etc
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
34
UC-6
Update the system's database
This use case is responsible for updating the system's database when new documents
have entered the medical document collection or the existing ones have been updated.
They are parsed, the document’s terms are extracted and (new) categories are
assigned.
Finally, the system updates or adds the corresponding indexes and the document to
the system’s database.
It is important to notice that for the updated articles the system keeps the same
authority score and the same idf’s (term frequency in the whole collection).
Goal in Context
The system processes the updated or the new documents from the medical document
collection and updates the system's database accordingly
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The re-crawling process has finished or new documents have entered
the medical document collection
Level: Summary
Scope : System Management Module
This use case is triggered by time or by Administrator
Triggering Action :
Flow of Events
Main Success Scenario:
1. The system gets a document (from the medical document collection) that has been
updated or added
2. The system converts the document into text and extracts its structural parts
3. The system extracts the document's terms, creates the corresponding
representation vectors and categorizes it to : (a) consumer - expert categories and
(b) Semantic Network MeSH categories (UC-8)
4. The administrator updates (or adds) document’s user-defined MeSH terms
5. The system assigns further categorization to the document
6. The system ensures that the document already exists in the system’s database and
performs the following:
6.1. The system creates or deletes the corresponding posting files for each of the
document’s representation vector that has been changed and saves the
updated document to the document's file keeping the existing document ID
6.2. The system creates or deletes the corresponding posting files for the
document’s authors, Medline terms and user defined MeSH terms that have
been changed
7. The system ensures that the document is new and performs the following:
7.1. The system builds the author indexes
7.2. The system ensures that the document is a research one by testing for the
existence of a bibliography part and builds the citation indexes (UC-7)
7.3. The system builds the indexes for the document's terms (lexicographic,
MeSH, UMLS) (UC-11)
8. The system gets the next document from the medical document collection that
has been updated or added and repeats steps 1-6 until all of them are updated or
added to the system's database
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
35
Flow of Events
Extensions:
4.a The Administrator doesn't assign new MeSH terms:
1. The system continues to step 5
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
36
UC-7
Build the citation indexes
This use case is triggered only for articles appeared in journals, workshop
proceedings, technical reports etc. First, the system normalizes the article’s
bibliography part by:
 converting the letters to uppercase and the hyphens to spaces,
 removing the citation tags at the beginning of the citation
 removing some words from a known pre-defined list such as 'in press',
'accepted for publication' etc and
 expanding abbreviations e.g. the word conf. to conference, proc. to
proceedings etc.
 normalizing the author names so that the misspellings and the name variants
are identified, e.g. the system :
- checks for first, last and middle names and
- converts the first and middle names to single letter initials
The normalization process is important as it helps the system to identify citations to
the same article written in different formats.
Afterwards the system processes the article's bibliography part and extracts (in a
tagged format e.g. XML) information about each of the article's references such as
title, information about the authors and information about the article's source e.g.
journal, book, conference etc, the date of publication and the context of each citation
from the body of the article.
Then the system creates the:
 forward links (see Glossary) to articles and
 backward links (see Glossary) from articles
that exist in the system's database. There are links associated with articles that do not
exist in the system's database yet. These links are added to a citation list (see
Glossary).
Finally the system builds the citation index which catalogues the citations that an
article makes, linking the articles with the cited works.
Goal in Context
The system processes the bibliography part of published documents and builds the
citation indexes.
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The document is an article (it has a bibliography part)
Level: User
Scope : Indexing subsystem
Triggering Action : The system performs normalization to the document’s
bibliography part.
Flow of Events
Main Success Scenario:
1. The system extracts information about the article's references
2. The system verifies that the cited article exists in the system's database
3. The system creates a forward link to this article
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
37
Flow of Events
4. The system repeats steps 2-3 until all the forward links of the article have been
created
5. The system verifies that there are citations to this article
6. The system creates the corresponding backward links
7. The system builds the citation indexes based on the article's forward and
backward links
8. The system creates the citation graph
Extensions:
2.a The cited article does not exist in the system's database:
1. Add information about the forward link to the citation list
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
38
UC-8
Extract the document's terms and categorize
the document
This use case extracts the document's terms and creates the document's representation
vectors (see UC-9 Create the representation vectors for the document). All the other
modules of the system mostly depend on the representation vectors produced here.
Then the system, based on the document's AMTEx terms, categorizes it to two
different types of categories: (1) consumer/expert categories with a weight of belief
that the document belongs to each category (see UC-10 Categorize the document to
consumer/expert category) and (2) the Semantic Network MeSH categories (see
Glossary).
Goal in Context
The system extracts the document's terms, creates three representation vectors and
categorizes it to: (a) consumer/expert categories, and (b) Semantic Network MeSH
categories
Details
Primary Actors: Administrator
Preconditions: The document has been converted to text (in case of articles this text
includes the title, abstract and the article's body)
Level: User
Scope : Document Analysis subsystem
The system processes the document
Triggering Action :
Flow of Events
Main Success Scenario:
1. The system extracts from the document the following terms: (a) the document's
lexicographic terms, which are mainly single word terms, (b) AMTEx terms and
(c) MMTx terms and creates the corresponding representation vectors (UC-9)
2. The system performs statistical terms analysis on the AMTEx terms
representation vector of the document and categorizes it (with a weight of
belief) to one of the consumer/expert categories (UC-10)
3. The system processes the AMTEx terms representation vector of the document,
extracts the Semantic Network MeSH categories and categorizes the document to
these categories
Extensions:
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
39
UC-9
Create the representation vectors for the
document or the user's query
This use case is responsible for creating the following representation vectors for the
document or the user's query:
(1) The dlex_terms, that contains the lexicographic terms, which are mainly single
word terms. Prior creating this vector, the system performs the standard
process in the information retrieval domain: it removes ''useless'' to the system
words e.g. stopwords, it reduces terms to their roots (stemming), it expands
acronyms, it merges multiple term entries for the medical text and finally it
compiles the within term frequency information.
(2) The dMMTx_terms, that contains MMTx (see Glossary) terms. These are UMLS
Metathesaurus terms(see Glossary).
(3) The dAMTEx_terms, that contains the AMTEx (see Glossary) terms. These are
MeSH terms.
Each of these vectors includes:
 the actual string (textual form of the terms found) and
 the term's frequency in the document or in the user's query.
Goal in Context
The system extracts terms from the document or the user's query and creates three
representation vectors
Details
Primary Actors: Administrator, Consumer, Expert
Supporting Actors:
Preconditions: The document's text has been extracted or the user has entered a query
Level: Subfunction
Scope : Document Analysis subsystem & Query Handling
subsystem
The system processes the user's query or a new document
Triggering Action :
from the medical document collection
Flow of Events
Main Success Scenario:
1. The system performs the standard process in the information retrieval domain and
creates the representation vector that consists of lexicographic terms
2. The system uses the AMTEx method and produces the representation vector of
AMTEx terms
3. The system uses the MMTx method and produces the representation vector of
MMTx terms
Extensions:
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
40
UC-10
Categorize the document to consumer/expert
category
This use case describes the process which categorizes a medical document as expert
or consumer and assigns a weight of belief that the document belongs to each of these
categories. During this process the AMTEx terms (see Glossary), which are actually
MeSH terms, are divided to one of the following categories:
1. consumer terms that belong both to Wordnet and MeSH (Wordnet∩MeSH),
2. expert terms that belong to MeSH but not to Wordnet (MeSH-Wordnet).
A document is considered to be an expert one, if it contains at least one expert term.
Otherwise it is considered to be a consumer.
# consumer _ terms
Furthermore, the document is assigned a value (which is the ratio
# total _ terms
# exp ert _ terms
for the consumer documents and the ratio
for the expert documents).
# total _ terms
These values are (its obvious that they are summed to one) are used later to improve
the system's retrieval performance (see UC-13 Retrieve relevant documents).
Goal in Context
The system categorizes the document as consumer or expert with a weight of belief
that it belongs to each category
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The AMTEx terms representation vector of the document has been
created
Level: Subfunction
Scope : Document Analysis subsystem
The system performs statistical term analysis on the AMTEx
Triggering Action :
terms representation vector of the document
Flow of Events
Main Success Scenario:
1. The system computes the number of consumer and the number of expert terms
2. The system ensures that the document has at least one expert term and categorizes
it as an expert one.
3. The system computes the weight of belief that the document belongs to the
consumer category
4. The system computes the weight of belief that the document belongs to the expert
category
Extensions:
2.a The document hasn't any expert terms:
1. The system categorizes it as a consumer document
Variations:
2. The system could categorize the document as an expert one, if it contains at least
2, 3 etc expert terms
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
41
UC-11
Build indexes for the document's terms
This use case is responsible for managing the indexing process based on the
document's terms, which are:
 the lexicographic terms,
 the AMTEx terms,
 the MMTx terms
 the user defined MeSH terms and
 the Medline terms (only for Medline documents).
During this process the system creates/updates four main data structures:
1. the Documents File which stores each document with a document ID,
2. a Vocabulary for each type of terms (lexicographic terms, MeSH terms, UMLS
terms) ,
3. the Inverted Index which stores the postings list for each of the terms mentioned
above (it also contains statistics on term frequency in each document) and
4. the Direct Index which associates a posting list for each document. This list
consists of:
 the document's AMTEx terms and
 the AMTEx term’s frequency in the document. The Direct Index can be used
to facilitate the query expansion process (see UC-16 Reformulate the user's
query) and the documents clustering process based on the AMTEx terms (see
step 10 UC-5 Create the system’s database).
Goal in Context
The system indexes the document by using Inverted Files.
Details
Primary Actors: Administrator
Supporting Actors:
Preconditions: The document’s representation vectors have been created
Level: User
Scope : Indexing subsystem
The system stores the document to the Document's File
Triggering Action :
Flow of Events
Main Success Scenario:
1. The system repeats steps 2-4 for each of the following terms : (a) lexicographic
terms, (b) AMTEx terms (c) MMTx terms (d) user-defined MeSH terms and (e)
Medline terms
2. The system adds all the distinct values of the representation vector to the
corresponding Vocabulary
3. The system creates the posting lists for the document's terms
4. The system verifies that the AMTEx terms representation vector is processed and
adds the document and the document's AMTEx terms to the Direct Index
Extensions:
1.a The document hasn't any user-defined MeSH or Medline terms :
1. The system doesn't repeat the steps 2-4
2. The system Exits
2.a There are terms that already exist in the Vocabulary:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
42
Flow of Events
1. The system continues to step 3
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
43
UC-12
Answer the query
The use case starts when the user submits a free text query. Consumers usually enter
long queries that describe their situation in great detail (e.g. medical history, family
medical history, where and how they feel uncomfortable, what happened the last
several days etc) in plain English - much like the way they talk to a doctor. The user
can also specify additional operations (and, or, not) to achieve more complex
queries such as:
 a particular query term should or should not appear in the retrieved
documents,
 two or more query terms should appear in the retrieved documents, and
 either one or another query term should appear in the retrieved documents.
Also, the user can switch between different types of representation for both the
document and the query ( UC-14, Rank the documents, lists all the combinations that
our system supports).
Prior handling the query the spelling corrector is called which corrects spelling
errors made while entering specialized, infrequently used medical terms in search
queries. The spelling corrector is valuable especially for the consumers who may
approach the medical domain in an approximate way.
Then the system retrieves
relevant documents in response to the user's query (see UC-13 Retrieve relevant
documents). Finally, the user views the results (see UC-15 Present results to user).
Usually he obtains a large list of documents or he had restricted so much the query
that the results are not sufficient. At this point he has to reformulate his query. The
system provides a variety of reformulating techniques in order to improve the retrieval
performance (see UC-16 Reformulate the user’s query).
Goal in Context
The user enters a query, retrieves relevant documents in response to his query and
reformulates the query if the results are inefficient.
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: The user is logged in the system
Level: User
Scope : User Interface Module
The user submits a free text query
Triggering Action :
Flow of Events
Main Success Scenario:
1. The user specifies additional operations (and, or, not) to achieve more complex
queries
2. The user selects the type of terms he desires for the retrieval process
3. The system ensures that the query hasn't any spelling errors
4. The system retrieves relevant documents in response to the query (UC-13)
5. The system presents the results to the user (UC-15)
6. The user reformulates his query (UC-16)
7. The System repeats steps 4-6 until the user indicates that he has finished
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
44
Flow of Events
Extensions:
1.a The user doesn't specify additional operators or values to metadata fields:
1. The system continues to step 2
3.a The query has spelling errors:
1. The system suggests the possible corrections
2. The user selects a suggestion or ignores them
7.a The user doesn’t select to reformulate his query:
1. The system Exits
Variations:
3. Various correction algorithms could be used, such as the Levenstein edit distance
algorithm, in order to create the list with the most similar candidates
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
45
UC-13
Retrieve relevant documents
This use case is responsible for the retrieval process. During this process:
(a) The system creates three representation vectors for the user's query (see UC-9
Create the representation vectors for the document or the user's query)
(b) The system (based on the Vector Space Model), retrieves a list of relevant
documents. A similarity score is assigned to each of the documents retrieved
by estimating the relevance between the document’s content and the user's
query (see UC-14 Rank the documents). The documents in the answer set are
ordered by this similarity score. All documents up to a prespecified degree of
similarity (defined by a query threshold) are retrieved. Alternatively the top k
(e.g. 20) documents are retrieved.
(c) Finally, the system allows the retrieved list of documents mentioned above, to
be altered to take into account additional types of evidence (see steps 3-5)
such as the user's type (consumer or expert) and the user’s favorite
categories(only for expert users). The filtering process is valuable in cases
where he wants to restrict the domain of the documents being retrieved.
Goal in Context
The system compares the user's query with each document from the database and
retrieves a list of relevant documents.
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: The user has logged in the system
Level: User
Scope : Query Handling subsystem
The user enters a non empty free text query
Triggering Action :
Flow of Events
Main Success Scenario:
1. The system extracts (a) the lexicographic terms, (b) the AMTEx terms and (c) the
MMTx terms from the query and creates the corresponding query's
representation vectors (UC-9)
2. The system computes a similarity score, determines the documents that match
the specific query and ranks the documents by this score(UC-14)
3. The system verifies that the user is a consumer and computes a score for each of
the documents retrieved by multiplying the document's similarity score with the
document's weight of belief that it belongs to the consumer's category
4. The system verifies that the user is an expert and computes a score for each of the
documents retrieved by multiplying the document's similarity score with the
document's weight of belief that it belongs to the expert's category
5. The system verifies that the user is an expert, and filters the re-ranked list of
documents by dropping the documents that do not belong to his favorite categories
Extensions:
5.a The expert hasn't any favorite categories or has disabled this option :
1. The system Exits
Variations:
2. The system could keep only the documents up to a prespecified degree of similarity
or the top k documents (where k is a user defined parameter)
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
46
UC-14
Rank the documents
This use case is responsible for computing the similarity score between each
document and the user's query, using the Vector Space Model (VSM). According to
this model the similarity score is computed as the sum of the inner products of the
document’s and the query’s representation vectors. It's up to the user to select (see
UC-12 Answer the query) from the following list which type of terms will be used for
the representation of both the documents and the query:
document
AMTEx terms(default)
MMTx terms
lexicographic terms
Medline terms(for Medline documents only)
Userdefined MeSH terms
query
AMTEx terms(default)
MMTx terms
lexicographic terms
AMTEx terms
AMTEx terms
Also, the system retrieves documents based on values in the author’s metadata field
Goal in Context
The system produces a ranked list of retrieved documents
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: The query's terms (lexicographic, MMTx and AMTEx terms) have
been extracted and the corresponding representation vectors have been created
Level: Subfunction
Scope : Query Handling subsystem
The system presents the user with a list of the different types
Triggering Action :
of terms that can be used for the retrieval process
Flow of Events
Main Success Scenario:
1. The system fetches the inverted lists for the query's terms
2. The system intersects them for conjunctive query or merges them for disjunctive
query
3. The system assigns a similarity score to each of the documents retrieved
4. The system ranks the retrieved documents according to this similarity score
Extensions:
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
47
UC-15
Present results to user
This use case is responsible for presenting the retrieved list of documents to the user.
The user has the option to change the ranking scheme and get the retrieved list sorted
by the document's (publication or creation) date or by the document's authority score
(see Glossary and UC-5 Create the system’s database step 9). Finally, the system
allows the user for navigation through the literature in time by following the links to
the articles that:
 the current document references and
 cite to the current document providing at the same time the context of each
citation i.e. the sentence(s) in which a reference is made to this article.
Goal in Context
The system presents the results to the user
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: A list of relevant documents to the user's query has been retrieved.
Level: Subfunction
Scope : User Interface Module
The system presents the total number of the relevant
Triggering Action :
documents retrieved in response to the user's query
Flow of Events
Main Success Scenario:
1. The system presents information for each document from the retrieved list such
as the document's title, the author name(s) etc
2. The system ensures that the document is an article, presents the document's
metadata i.e. publication date, lists out information about articles that are being
referenced by the current article (references), articles that cite to the current article
(citations) and the context of each citation
3. The system ensures that the document is an unstructured medical text and presents
the date the document was created or last updated and a link to the original URL
4. The user selects from the following list of options and changes the ranking
scheme :
4.1. Order the results by descending date (publication date for articles, creation for
unstructured medical texts)
4.2. Order the results by the document's authority score
5. The User selects a document and views extensive information
6. The system repeats steps 4-5 until the user indicates that he has finished
Extensions:
4.a The user changes the ranking scheme:
1. The system re-ranks the retrieved list of documents according to the user's
selected criteria
2. The system continues to step 1
5.a The document is an article:
1. The system indicates the article’s references and citations that exist in the
system's database
2. The user navigates through these articles by following the links (backward
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
48
Flow of Events
or forward)
Variations:
1. The system can present as answers to the query the top r documents, where r is a
parameter set by the user
5. The user might select to print, save or e-mail the document
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
49
UC-16
Reformulate the user's query
Often the retrieved documents are relevant to the query, but useless to the searcher. In
order to improve the search results, our system interacts with the user offering him
suggestions to add precision to his query. This is very valuable especially in the
medical field where consumers don't have the knowledge or vocabulary to specify
what really they want. Often, they need to retrieve relevant documents which may not
contain any occurrences of the original user's query terms. This use case describes the
query expansion process in order to help the users refine their queries. The query is
enriched with related terms such as :
 MeSH terms which are automatically extracted from the top-ranked
documents using the AMTEx method (see Glossary). Alternatively, the user
can suggest the documents that he considers to be relevant and the MeSH
terms can be extracted from these documents. The presence of the Direct
Index (see UC-11 Build indexes for the document's terms step 4), makes this
expansion efficient.
 MeSH terms in the neighbourhood of a query term (i.e. terms lower or higher
in the MeSH hierarchy) using a similarity threshold between 0 and 1. High
values of this threshold are desirable for very specific queries (in this case few
terms from the MeSH hierarchy are used for query expansion) while lower
values are desirable for too general queries where many documents can be
retrieved. Notice though that very low values for the threshold might de-focus
the query.
 synonyms (entry terms) from the MeSH hierarchy, e.g. 'pain' and 'ache' are
synonyms
Goal in Context
The user's query is expanded with new terms
Details
Primary Actors: Consumer, Expert
Supporting Actors:
Preconditions: The user has entered a query and retrieved relevant documents in
response to this query
Level: User
Scope : User Interface Module
The system provides the user with a list of options
Triggering Action :
Flow of Events
Main Success Scenario:
1. The user has the following list of options to select the source of the terms that
will be used for the query expansion:
1.1. AMTEx terms
1.1.1. The system extracts the MeSH terms from the top r retrieved
documents using the AMTEx method
1.1.2. The system performs frequency analysis and a frequency score is
associated to each term
1.1.3. The system ranks these terms according to frequency
1.1.4. The system presents the terms in order of decreasing frequency and a
brief explanation for each MeSH term e.g. display the 'Scope Note' field
from MeSH to inform the users, especially the consumers.
1.2. MeSH terms from the MeSH hierarchy using a similarity threshold
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
50
Flow of Events
1.3. synonyms (entry terms) from the MeSH hierarchy
2. The user selects the list of terms that interest him
3. The system expands the query with these new terms
4. The system re-weights the expanded query
Extensions:
Variations:
1.1.1a The user could select which of the retrieved documents are relevant and the
system extracts the MeSH terms from these documents
1.1.1b The user could select the number r of the top-ranked retrieved documents (the
system Administrator defines a default value) that will be used to extract the MeSH
terms for the query expansion
1.2. The user can update the value of the similarity threshold.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
51
UC-17
Browse database by medical topics
This use case allows the expert to browse the system's database documents by medical
topics. Our system can support three different types of categorization for the
documents: the Semantic Network MeSH categories ,categories produced during the
clustering process (based on the Citation Graph) and categories produced during the
document clustering process.
The user can switch between these different types of categories and retrieve different
results.
Goal in Context
The system allows the expert to view the documents by browsing the medical topics
Details
Primary Actors: Expert
Supporting Actors:
Preconditions: The expert is logged on
Level: User
Scope : Browsing subsystem
The expert selects the type of categories he wants to use for
Triggering Action :
browsing
Flow of Events
Main Success Scenario:
1. The system presents the medical topics
2. The user selects the topics that interest him
3. The system retrieves all the documents that belong to this topic (and to the topic’s
children etc)
4. The system computes a score for each of the documents retrieved by multiplying
the document's relevance score with the document's weight of belief that it
belongs to the expert's category
5. The system presents the results to the user (UC-15)
6. The system repeats steps 1-5 until the user indicates that he has finished with the
browsing process
Extensions:
Variations:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
52
3.4 ACTIVITY DIAGRAMS
The activity diagrams are used to model a use case scenario in great detail. Each activity
diagram starts with a black circle and stops at (at least one) concentric white/black circles. The
actions are eclipses. When an action is decomposed into sub-actions, boxes with rounded ends
are used. From each action comes out an arrow, called transition, connecting it to the next one.
The arrow can also lead to a diamond, which indicates branching into two or more mutually
exclusive transitions. Expressions inside brackets [ ], e.g. [username exists] or [username
doesn’t exist], label the transitions coming out of a branch. Also a transition might fork into
two or more parallel activities e.g. the activities ‘Categorize the document to consumer/expert
category’ and ‘Categorize the document to Semantic Network MeSH categories’ (see Activity
Diagram#8) can be performed in parallel. When both activities have finished the flow
continues. Each fork must have a subsequent join. Forks and joins both appear in the activity
diagram as solid bars.
The seventeen activities diagrams (one for each of the use cases described above) follow:
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
53
A
A
[continue]
Figure 3.5: Activity diagram#1 – Use the application
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
54
Figure 3.6: Activity diagram#2 – User registration
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
55
Figure 3.7: Activity diagram#3 – User login
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
56
Figure 3.8: Activity diagram#4 – System management
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
57
A
A
Figure 3.9: Activity diagram#5 – Create the System’s database
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
58
A
A
Figure 3.10: Activity diagram#6 – Update the System’s database
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
59
A
A
Figure 3.11: Activity diagram#7 – Build citation indexes
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
60
Figure 3.12: Activity diagram#8 – Extract the document’s terms and categorize the document
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
61
Figure 3.13: Activity diagram#9 – Create the representation vectors for the document or the user’s query
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
62
Figure 3.14: Activity diagram#10 – Categorize the document to consumer/expert category
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
63
A
A
Figure 3.15: Activity diagram#11 – Build indexes for the document’s terms
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
64
Figure 3.16: Activity diagram#12 – Answer the query
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
65
Figure 3.17: Activity diagram#13 – Retrieve relevant documents
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
66
Figure 3.18: Activity diagram#14 – Rank the documents
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
67
Figure 3.19: Activity diagram#15 – Present results to user
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
68
Figure 3.20: Activity diagram#16 – Reformulate the user’s query
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
69
Figure 3.21: Activity diagram#17 – Browse database by medical topics
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
70
3.5 CLASS DIAGRAMS
Class diagrams describe the classes inside the system and the relationships between them. As the
proposed system is large, the system’s classes are organized into packages (see Figure 3.1- package
diagram).
The elements of a class diagram that this work uses are:
 A class is drawn as a rectangle with three compartments. The first describes the class’s name, For
naming the class two rules have been followed [6]: (a) the vocabulary of the medical domain is used
and (b) all names are singular nouns starting with a capital letter. The second compartment displays the
class’s attributes, Here usually only the single valued attributes are described. The multi-valued or the
complex attributes form new classes and an association link between the classes is shown e.g. the
classes Date, Address, Phone (see Figure 3.22 - System Management Package-class diagram). Finally,
the class’s operations are displayed in the third compartment and represent the behaviour of the class.
When a class is imported by another package, then only the class’s name and namespace is displayed
(e.g.
see
Figure
3.25
-
User
Interface
Package-class
diagram,
where
the
class
User_Management_Package::loginManager is imported).
 The system’s class diagrams have three kinds of relationships:
- association which is the simplest type of relationship between the instances of two classes, shown
as a uni or bi directional line connecting the two classes,
- aggregation is between the whole and its parts and is a stronger relationship than association. It is
shown as a line with a white diamond next to the class representing the whole e.g. the class
SemanticData(see Figure 3.22 - System Management Package-class diagram) consists of a
collection of Medline_terms, AMTEx_terms etc
- composition which is a stronger relationship than aggregation. Here the whole has responsibility
for the existence of its parts, e.g. if parsedQuery (see Figure 3.23 – Document Retrieval Packageclass diagram) is deleted then MMTx_terms, AMTEx_terms etc are also deleted.
The relationships mentioned above have multiplicity numbers on their edges (they define how many
objects participate in this relationship).
Also the system’s class diagrams have inheritance relationships (one class, the child, inherits the
functionality of another class, super class, and then adds new functionality of its own). To model
inheritance on a class diagram, a solid line is drawn from the child class with a triangle pointing to the
super class (see Figure 3.22 - System Management Package-class diagram) where an Expert is a
registeredUser with extra functionality.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
71
3.5.1 CLASS DIAGRAM#1 – SYSTEM MANAGEMENT PACKAGE
Here follows a brief description of the most important classes of the System’s Management Package
(see Figure 3.22):
Class Name
Description
MedicalDocumentCollection This class models a set of medical documents which
is
either
a
Medical_Database
or
a
CrawledCollection.
Document
This class encapsulates the most fundamental
concept to indexing which is a medical document. It
is essentially an iterator over the documents in the
MedicalDocumentCollection
SemanticData
This class encapsulates the central aspects that can be
found in the document’s contents as well as other
significant information discussed in the document.
These are a collection of:
Userdefined_MeSH_terms,
Medline_terms,AMTEx_terms and MMTx_terms.
Descriptive_Data
This class encapsulates the document’s metadata
such as the AuthorInfo, and the document’s
publication Date.
LexicographicTerm
This class models the document’s lexicographic
terms.
Article
This class encapsulates another fundamental concept
for indexing which is a medical article. Each article
has a collection of CitationMetadata, belongs to a
Categories (Category) and is published to a source
(SourceInfo) by a publisher (PublisherInfo).
User
This class models the different types of users. Each
user is either the Administrator or a registeredUser
(Consumer or Expert).
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
72
*
TECHNICAL UNIVERSITY
OF CRETE
C THESIS
Figure
3.22:– MS
System
Management
Package – Class diagram
73
3.5.2 CLASS DIAGRAM#2 – DOCUMENT RETRIEVAL PACKAGE
Here follows a brief description of the most important classes of the Document Retrieval Package (see
Figure 3.23):
Class Name
parsedQuery
Description
This class encapsulates the query terms that will be
used for matching the query with the stored
documents. These terms are a collection of:
Lexicographic_terms,
MMTx_terms
and
AMTEx_terms
expandedQuery
This class encapsulates the query terms after
expansion
with
new
AMTEx_terms,
MeSH_hierarchy_terms or synonym_terms
resultSet
This class models a result set of documents in respect
to the user’s query to a medical Category (selected
by the user). The result set is a collection of
documents (DocumentRetrieved).
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
74
Figure 3.23 : Document Retrieval Package – Class diagram
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
75
3.5.3 CLASS DIAGRAM#3 – USER MANAGEMENT PACKAGE
Here follows a brief description of the most important classes of the User Management Package (see
Figure 3.24).
Class Name
Description
loginManager
This class is responsible for user’s login
registrationManager
This class is responsible for user’s registration
Figure 3.24: User Management Package – Class diagram
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
76
3.5.4 CLASS DIAGRAM#4 – USER INTERFACE PACKAGE
Here follows a brief description of the most important classes of the User Interface Package (see Figure
3.25):
Class Name
Description
registrationForm
It encapsulates the concept of the user.
query
It models the user’s natural language query.
results
It models the retrieved documents in response to the
user’s query or in response to the browsing
Category. This set is a collection of elements (the
resultElements) which are either medical
Web_pages or Articles.
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
77
Figure 3.25 : User Interface Package – Class diagram
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
78
Chapter 4
IMPLEMENTATION
To demonstrate and objectively asses the quality of the design, a demonstrator medical information
system is implemented. This implementation doesn’t support the full range of functionalities presented
in the requirements analysis phase (section 3):
-
The example medical document collection (see Glossary) consists of around 30,000 references from
the OHSUMED collection. OHSUMED is a MEDLINE (see Glossary) subset, consisting of 348,566
references, covering all references from 270 medical journals over the period 1987-1991. The
available fields are title, abstract, MESH indexing terms (in this thesis called MEDLINE terms),
author, source and publication type. Most references are accompanied by abstracts, while others are
not.
-
Lexicographic, AMTEx terms (MeSH terms extracted from the title and abstract) and MeSH terms
assigned to documents by domain experts (already available with the documents) are used for
indexing.
-
Each document if further categorized to one or more Semantic Network categories (see Glossary).
The additional types of medical categories described in the requirements analysis phase (categories
created during the document clustering process or during the citation graph Link Clustering process)
are not supported yet.
The following use cases (or part of them) have been implemented:

(UC-2) User registration

(UC-5) User login

(UC-8) Extract the document's terms and categorize the document

(UC-9) Create the representation vectors for the document or the user's query. Only the
representation vectors for the lexicographic and AMTEx terms are created.

(UC-10) Categorize document to consumer/expert category

(UC-11) Build indexes for the document's terms. The first version doesn’t perform indexing based
on user defined MeSH terms and on MMTx terms.

(UC-13) Retrieve relevant documents. This version maps the query to AMTEx terms only. Also in
this version the expert’s favorite categories are ignored e.g. they are not used to improve ranking in
the answer set.

(UC-14) Rank the documents
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
79

(UC-15) Present results to user. Most of this use case’s functionality is not supported. Extensive
information documents are not available (OHSUMED documents have only abstracts). The
retrieved answers to user queries contain information such as the document’s title, abstract (if it
exists), document’s categories (from the Semantic Network MeSH hierarchy) and the PubMED
identifier.

(UC-17) Browse database by medical topics. This system supports only topics from the Semantic
Network MeSH hierarchy.
For the text indexing and retrieval process Apache Lucene library is used. Lucene is a free source
search engine library written entirely in Java, supported by the Apache Software Foundation. The
following figures illustrate systems interface for user registration, query expression and results
presentation.
Figure 4.1: User registers into the system
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
80
Figure 4.2: User submits a free text query
Figure 4.3: Results presentation
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
81
Glossary
Term
AMTEx
Meaning
AMTEx is a method that automatically extracts MeSH terms
from medical text.
authority score
This score is computed during the Citation Graph Link Analysis
process. It characterizes the article's popularity and is used for
the ranking process.
backward links
These are directed links from a list of articles to one article and
imply that the first articles cite to the second one.
citation graph
Is a directed graph where each node represents an article and a
directed link from one node to another node implies that the
article associated with the first node cites the article associated
with the second node.
citation list
This list consists of the forward and backward links of the
system's database articles that reference to or are cited by articles
that do not exist in the system's database yet.
forward links
These are directed links from one article to a list of other
articles and imply that the first article has references to all the
other articles.
HON accredited sites
These are sites that are accredited from the Health On the Net
Foundation (HON). HON is a non Governmental Organization
whose mission is to 'guide Internet users by highlighting
reliable, understandable, relevant and trustworthy sources of
online health and medical information'. Each HON accredited
site is re-evaluated every year.
medical document
collection
It consists of documents that belong either to existing medical
databases (such as Medline) or to Web’s HON accredited sites.
MEDLINE
It is a large medical collection. Each document contains semistructured information in XML, has a unique PubMed id,
includes citation information such as author(s), journal, year etc.,
has manually assigned MeSH terms, called Medline terms etc
MeSH
MeSH (Medical Subject Headings) is a taxonomic hierarchy of
medical and biomedical terms suggested by the U.S. National
Library of Medicine.
MMTx
MMTx (MetaMap Transfer) is a tool from the U.S. National
Library of Medicine, that maps biomedical text into concepts in
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
82
Term
Meaning
the UMLS Metathesaurus.
Semantic Network
MeSH categories
It is part of the National Library of Medicine's (NLM)
Semantic Network(SN). It consists of :
(1) the 135 Semantic Types of the SN which provide a consistent
categorization of the MeSH terms only (SN categorizes all terms
represented in the UMLS Metathesaurus) and
(2) only the 'isa' link from the Semantic Relations that exist
between the Semantic Types, establishing a hierarchy of types
within the SN.
system's database
It consists of the documents from the medical document
collection which are analyzed, indexed and stored.
UMLS Metathesaurus
The UMLS (Unified Medical Language System) Metathesaurus
includes over 100 incorporated controlled vocabularies (such as
SNOMED CT, MeSH, ICD-9-CM etc).
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS
83
References
1. Rumbaugh J, Jacobson I, Booch G. The Unified Modelling Language Reference Manual,
Addison Wesley Longman Inc, 1999
2. Fowler Martin, UML Distilled Third Edition A Brief Guide to the standard Object Modelling
Language, Addison-Wesley, 2003
3. Hliaoutakis A, Zervanou K, Petrakis E, Medical Document Indexing and Retrieval: AMTEx vs.
NLM MMTx , Proceedings of the 12th International Symposium on Health Information Management
Research 2007
4. Gaudinat A, Ruch P, Joubert M, Uziel P, Strauss A, Thonnet M, Baud R, Spahni S, Weber P,
Bonal J, Boyer C, Fieschi M, Geissbuhler A. Health search engine with e-document analysis for
reliable search results. International Journal of Medical Informatics 2006 Jan;75(1):73-85
5. Cockburn Alistair, Writing Effective Use Cases, Addison Wesley, 2001
6. Christodoulakis S., Notes from the course ‘Ανάπτυξη εφαρμογών πληροφοριακών συστημάτων
στο Διαδίκτυο’ Chapter 7, 2007
7. Object Management Group - Unified Modeling Language [2008 January]; www.uml.org
8. MeSH: The Medical Subject Headings. [2008 January]; http://www.nlm.nih.gov/mesh
9. Unified Medical Language System (UMLS). [2008 February];
http://www.nlm.nih.gov/research/umls/
10. MetaMap Transfer (MMTx). [2007 March]; http://mmtx.nlm.nih.gov
11. Online Indexing Training Module [2007 June]; http://www.nlm.nih.gov/bsd/indexing
12. Hliaoutakis A, Zervanou K, Petrakis E, Milios E, Automatic Document Indexing in Large
Medical Collections, ACM International Workshop on Health Information and Knowledge
Management (HIKM 2006), November 11, 2006, Arlington, VA, US
13. C. Lee Giles , Kurt D. Bollacker , Steve Lawrence, CiteSeer: an automatic citation indexing
system, Proceedings of the third ACM conference on Digital libraries, p.89-98, June 23-26, 1998,
Pittsburgh, Pennsylvania, United States
14. Huajing Li, Isaac G. Councill, Levent Bolelli, Ding Zhou, Yang Song, Wang-Chien Lee, Anand
Sivasubramaniam, C. Lee Giles, ‘CiteSeerχ – A Scalable Autonomous Scientific Digital Library,
ACM International Conference Proceeding Series; Vol. 152, 2006, Hong Kong
15. U.S. Department of Health and Human Services, Office of Disease Prevention and Health
Promotion,’Expanding the reach and impact of consumer e-health tools’, June 2006
16. OHSUMED Test Collection ir.ohsu.edu/ohsumed/ohsumed.html
TECHNICAL UNIVERSITY OF CRETE – MSC THESIS